Novel Fingerprint Liveness Detection with Fractional Energy of Cosine Transformed Fingerprint Images and Machine Learning Classifiers

Abstract

Biometric identification has gained more popularity and trust because of the ease of accessibility and unique person identification. Fingerprints are most widely used and studied in biometric identification. These days impersonation of fingerprints by means of gelatin, wood glue , cadaver are posing new challenges for fingerprint based biometric identification system. Hence fingerprint liveness detection has become one of the primitive and essential research areas .Machine learning classifiers may help in fingerprint liveness detection with the help of properly extracted features of fingerprints. The orthogonal transforms may support proper extraction of features because of energy compaction capability, for swifter and more accurate fingerprint liveness detection with fractional coefficients considered as feature vectors .The paper propose novel fingerprints liveness detection techniques with fractional coefficients of cosine transformed fingerprints samples and machine learning classifiers .Experimentation is carried out with eight propositions of fractional coefficients of Cosine transformed fingerprints considered to form feature vectors with four assorted machine learning classifiers and tested on 2 benchmark datasets ATVS and FVC 2000 .The classification accuracy is used to compare the performances of the variations of proposed fingerprint liveness detection method .Overall better liveness detection is observed with 0.094%of fractional coefficients for random forest classifiers closely followed of 0.094 fractional coefficients .The feature level fusion of 0.094% and 0.024% of fractional coefficients has given further boost in accuracy of fingerprint liveness detection.